Ray Tracing Vision for Video Game Characters

1
Ray Tracing Vision for Video Game Characters

• Alex JerezThesis Defense2004

2
Overview

• Introduction
• Key Elements
• Ray Tracing vs Checking Distance
• A. I.
• Research Results
• Conclusions and Future Work

3
Introduction

• Although video games have become more and more
  lifelike there are parts of the video game that
  are still hard to simulate. Character vision is
  an example of one of these simulations.
• A player sees the world from either a third
  person or first person perspective, but this
  simulates little about how our vision works.

4

• Characters tend to have a set of rules to follow,
  called a knowledge base, that allows them to
  perform certain operations but they are generally
  not able to learn from their environment and the
  objects surrounding them.
• This thesis then focuses on the later idea and
  tries to answer the following questions

5

• What happens when a character has no knowledge of
  the environment in which it is placed but that it
  is given a set of basic rules by which it can
  learn to interact with objects and other
  characters?
• What if characters and elements are independent
  from each other until the moment that they become
  aware of each other existence?
• How well can a character behave when faced with
  an environment that is similar to one from where
  he has gotten previous experience?

6

• The purpose of this thesis is then to create a
  character that behaves not based on given
  knowledge of all the elements in the world.
  Placing the character in a grid with cells then
  we want our character to move from point A to
  point B while trying not to be caught.

7

• The characters behavior is based on what it is
  surrounded at certain time during its journey.
• The discovery of a new objects using its own
  vision

8
Key Elements

• Spatial Data structure Quadtree
• Objects and Characters
• Physical Animation and Key Framing
• Ray Tracing
• A.I

9
Quadtree

• Easy way to find the location of a point

10
Objects

• All objects and characters are meshes so that we
  can have more control over the vertices.
• There are three different kinds of objects
  Spheres, Boxes, and Cylinders.
• There is a main character and enemy characters
  trying to prevent the main character from
  reaching his goal.

11
Animation

• Objects are animated using physics for rigid
  bodies.

12
Animation

• Characters are animated by borrowing the key
  framing technique from 2D animation.

Frame 0
Frame 5 Frame 10
time
13
Collisions
a)
Distance
b)
a) Case in which the distance between two objects
is greater than the sum of both radius. In this
case there is no collision. b) Collision occurs
in this case since the distance is not greater
than the sum of the radiuses. The direction
vectors are calculated and the objects will
bounce each in their own direction.
14
Ray Tracing Camera Model

• Models the way in which light rays react with
  surfaces to produce color.
• We are not interested on finding the color of
  surfaces, we are interested in finding which
  objects can be seen by the character.
• Ray tracing is computationally expensive, for
  this project we are using just a small number of
  samples to have an idea of what the character is
  looking at.

15
Checking the distance to all the objects

16
Checking the field of vision of the character

17
Camera model
18

• Low number of samples
• Medium number of samples
• High number of samples

19

• With this camera model we can recognize which
  objects are in the field of vision of the
  character.

Look at Map
20
How can we use this information?

• The character can scan cells and decide which
  ones are safe to move to.
• If the character finds objects on his path he can
  decide whether to push them out of the way or use
  them to stall other characters.
• The character can recognize whether he is in a
  safe place where no characters are around him or
  if he is in danger he will move to close to them.

21
A.I.

• All the information required for a character to
  react is stored in what is called the Knowledge
  Base. This database contains all the required
  data structures to allow the character to make
  decisions.
• The simplest way to move from point A to point B
  is a straight line, but this idea assumes that
  there is nothing in the path of the character.

22
Searching Techniques

• Characters use 3 different types of searching
  techniques
• Bread First Search (BFS)
• Hill Climbing (HC)
• A Star Search (A)

23
BFS

• When a BFS is assigned to the character, he will
  search the cells using an expanding circle
  method.

24

• Search tree created using BFS

25
HC

• Instead of searching the cells level by level, we
  are going to search for those cells which have a
  better chance of letting our character reach its
  goal. We use a state function to decide.
• The state function is constructed using
• distance from the current cell to the goal cell
• a penalty if there are any enemies in the cell.

26
(No Transcript)
27

28
A Search

• So far the character does not keep track of how
  much he has had to walk or what probability
  certain cell has to contain enemies.
• We can exploit this information and store it into
  the characters knowledge base so that it can be
  used again when the character is placed again in
  a similar scenario.

29

• When the character sweeps the cells around the
  one where he is standing at, he will use the same
  function as the HC function to calculate the
  state value of the cell, but in addition the
  character will also add the distance traveled so
  far, and a probability of the cell containing
  enemies.
• The first time the character is placed within a
  scenario he stores the additional information.
  When he is placed in a similar environment the
  character can use this information to make better
  decisions.

30
a)
b)

c)
a) Original HC algorithm solution b) Distance
walked so far each time the characters moves
from cell to cell. c) Probability of finding
enemies in the cell. For each character found in
a cell the probability goes .25 higher
31

• What happens the next time the character is faced
  with a similar scenario?

32
Decision Trees

• A decision tree is defined as a structure that
  takes as input an object or situation described
  by a set of properties, and outputs a yes/no
  decision. Decision trees therefore represent
  Boolean functions or functions with a large range
  of outputs

33
Decision Tree for an Enemy Character
34
Decision Tree for the main Character
35

36
Conclusions

• What happens when a character has no knowledge of
  the environment in which it is placed but that it
  is given a set of basic rules by which it can
  learn to interact with objects and other
  characters?
• Implementation generates on characters the
  element of surprise
• The vision seemed to generate a very accurate
  approximation of the many objects in the
  characters field of vision.
• The character reaction with objects is also
  fluent and the characters set of rules to deal
  with them seem to be everything needed for the
  character to solve the spatial problems.
• The only problem with using ray tracing is that
  the rays shot from the eye might miss some
  objects in the world depending on the samples
  that we are taking.

37

• What if characters and elements are independent
  from each other until the moment that they become
  aware of each other existence?
• Computationally objects are not independent of
  each other.
• Instead, we achieve this independency by letting
  the characters use a discovery method based on
  placing the character at some point and then make
  decisions depending on what it sees around him.
• This system seems to allow character to choose
  helpful cells preventing him from moving to what
  looks as a crowded cell
• Enemies using a BFS decision trees were best fit
  to guard cells.

38

• The Hill Climbing and A algorithms did not work
  too well with the enemies but worked great with
  the main character.
• The best combination for which the system seem to
  have the better balance is to assign a BFS to the
  enemies and the A to the main character. The HC
  search produced a good behavior but it was a
  little bit too static for a system that is always
  moving.

39

• How well can a character behave when faced with
  an environment that is similar to one from where
  he has gotten previous experience?
• The system showed that the A search was the
  search method that will yield more information
  that can be used in similar environments.
• The BFS has no useful information especially
  because it is a sequential search.
• The HC search has some useful information but it
  is still using the number of characters found
  using the look() function and this dependency
  makes one system different from the other even if
  they are started exactly the same.
• The A system with the saved probabilities works
  better for a similar environment since it is not
  dependent only in how many enemies can be seen
  but also in the probability of finding enemies on
  that cell.

40
Future Work

• Extend the ability for the characters to move in
  a more complex environment with hills and
  valleys.
• The character animation can be studied further by
  adding more variable values to the character,
  such as stamina and noise made while walking.
• The optimization of the searching algorithms can
  be studied further by finding better evaluation
  functions when computing the state value of a
  cell. (character speed, distance needed to walk
  and the history of characters seen so far)

41
Acknowledgments

• Dr. James C. Miller
• Dr. Christos Nikolopoulos
• Dr. Jian Liu

42
Bibliography

• 1 Herman, Leonard, Jer Horwitz, Steven Kent
  and Skyler Miller. The
• History of Video Games. Gamespot. 2002. www.gamespot.com
• 2 Wedge vs. Needle. Emulators Unlimited. 2001.
• 
  
• 3 A History of Video Games. Geek Comix. 2002.
• 
  
• 4 Bells, Mary. Computer and Video Game History
  About.com 2000
• a090198.htm
• 5 Pokorny, Cornel K., and Curtis F. Gerald.
  Computer Graphics The
• Principles Behind the Art and Science. Franklin,
  Beedle Associates. 1989.
• 6 Ferraris, Jonathan. Quadtrees. Game
  Developers. 2004
• /features/quadtrees

43

• 7 Hunter, William. The History of Video Games
  From Pong to
• Pac-man. Design Boom. 2000.
• ng.html
• 8 Hill, F.S. Jr. Computer Graphics Using
  Opengl. Second Edition
• Prantice Hill 2001.
• 9 Witkin, Andrew and David Baraff. Physically
  Based Modeling.
• Online Siggraph 2001. Course Notes
• pbm2001/
• 10 Moment of Inertia. Wolfram Research. 2004.
• 11 Dopertchouk, Oleg. Simple Bounding-Sphere
  Collision Detection.
• Game Developers. 2001.
• s/article1234.asp
• 12 Kelly, Doug. Character Animation in Depth.
  The Coriolis Group Inc.
• 1998. (p.p. 616)
• 13 Watt, Alan. 3D Computer Graphics. Third
  Edition.
• Addison-Wesley Publishing Company. 2000.

44

• 14 Nikolopoulos, Chris. Expert Systems
  Introduction to First and
• Second Generation and Hybrid Knowledge Based
  Systems. Marcel Dekker Inc. 1997
• 15 Ellis Horowits, Sartaj Sahni, Sanguthervar
  Rajasekaran. Computer
• Algorithms C. Computer Science Press.
  1998.
• 16 Cormen, Thomas, Charles Leiserson, and
  Ronald Rivest.
• Introduction to Algorithms. The MIT
  Press. 2000.
• 17 Luger, George. Artificial Intelligence
  Structures and Strategies
• for Complex Problem-Solving. Third
  Edition. Addison-Wesley.
• Publishing Company..1997.
• 18 Russell, Stuart and Peter Norvig. Artificial
  Intelligence A Modern
• Approach. Prentice Hall. 1995.
• Other Consulted Bibliography.
• Foley, James, Andries van Dam, Steven Feiner,
  John Hughes.
• Computer Graphics Principles and Practice.
  Second Edition. Addison-Westly Publishing
  Company. 1997.
• Woo, Mason, Jackied Neider, Tom Davis, and Dave
  Shreiner.
• Opengl Programming. Third Edition. Addison-Westly
  Publishing Company. 1997.
• Hawkins, Kevin, Dave Astle, and Andre LaMothe.
  OpenGL Game Programming. The Premier Press. 2004.
• Shirley, P and Peter Shirley. Fundamentals of
  Computer Graphics. A.K. Peters Ltd. 2002.